Remove yolo-face and use standard YOLO library only (optimization)

[spacegoatai]

Background

The current yolo-face model (yolov11m-face.pt from https://github.com/akanametov/yolo-face) includes CLIP-ViT-B-32 which is loaded on every image processing cycle, causing performance overhead.

Details

• The yolo-face model loads CLIP-ViT-B-32-laion2B-s34B-b79K internally
• This happens because yolo-face includes detection for faces, drones, and football players
• We only use it for face detection
• The standard YOLO model can detect faces reasonably well

Proposed Change

Remove the custom yolo-face model and use the standard YOLO model for face detection. This would:

• Eliminate the CLIP model loading overhead
• Simplify the codebase
• Slightly reduce face detection accuracy (but standard YOLO is still quite good)

Files to Update

• kalliste/config.py - Remove yolo-face model configuration
• kalliste/detectors/yolo_face_detector.py - Could be removed or refactored
• kalliste/detectors/detection_pipeline.py - Update to use standard YOLO for faces
• kalliste/model/model_download_manager.py - Remove yolo-face download
• kalliste/model/model_registry.py - Remove yolo-face initialization

Note

This is marked as an optional optimization. The current setup works well, but removing the CLIP dependency would improve processing speed.